Steamhammer plans
Time for another update on my plans, since they change almost as fast as I can think up changes. After the big push for AIIDE 2020 I have been working slowly, but I’m still making progress.
I’ll release a version 3.2.20 soonish, with small bug fixes and other minor tweaks. For example, I found the cause of failed macro hatchery conversion, and I’m correcting it now. None of the fixes should make a big difference in playing strength, but I know from experience that many small improvements add up to a big improvement. Next, if I can get it to work this time, will be version 3.3 on BWAPI 4.4.0, a necessary upgrade. I expect I will be unable to resist features enabled by 4.4.0, like delaying overlord speed until after hive and tracking enemy terran scans.
Then I want to at long last make visible progress on my main objective of strategy adaptation. I’ve changed my mind repeatedly about what the next step should be, and the result is that I have nonworking partial implementations of several complex features. I don’t see it as wasted work, because all the features are necessary, but it does mean I have no progress to show off. I want to bite off a piece small enough that I can swallow it down in time for the year-end SSCAIT tournament. I think it should be either the opening timing stuff that came up yesterday (it does give useful though ambivalent hints), or else some machine learning stuff I’ve mentioned earlier, whose code has been sitting in an almost-finished state, not yet passing small-scale tests and unready to be integrated into the whole program for its big test. Whether I make it in time for SSCAIT or not, I expect my choice to be the theme of the 3.4.x series.
Next: Specifics on opening timing.
Trackbacks
No Trackbacks
Comments